radeonsi: optimize out the loop in si_get_ps_input_cntl

Use a remap table from a semantic to an index instead of searching
for the correct index.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6340>

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 7ad282a651f1c5a4d5c3101f53ac8d91d3c0167e..8694bc58ab71469aecf4980ca196de9412c5684d 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -327,6 +327,7 @@ struct si_shader_info {
    ubyte input_interpolate[PIPE_MAX_SHADER_INPUTS];
    ubyte input_usage_mask[PIPE_MAX_SHADER_INPUTS];
    ubyte output_semantic[PIPE_MAX_SHADER_OUTPUTS];
+   char output_semantic_to_slot[VARYING_SLOT_TESS_MAX];
    ubyte output_usagemask[PIPE_MAX_SHADER_OUTPUTS];
    ubyte output_readmask[PIPE_MAX_SHADER_OUTPUTS];
    ubyte output_streams[PIPE_MAX_SHADER_OUTPUTS];

diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c b/src/gallium/drivers/radeonsi/si_shader_nir.c
index 1cf3f591248f9becd87972ca4306cd1212e9d034..a2f3af831ad8e2596c758d9de4643de8ea9f4471 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
@@ -143,12 +143,14 @@ static void scan_io_usage(struct si_shader_info *info, nir_intrinsic_instr *intr
    } else {
       /* Outputs. */
       assert(driver_location + num_slots <= ARRAY_SIZE(info->output_usagemask));
+      assert(semantic + num_slots < ARRAY_SIZE(info->output_semantic_to_slot));
 
       for (unsigned i = 0; i < num_slots; i++) {
          unsigned loc = driver_location + i;
          unsigned slot_mask = (dual_slot && i % 2 ? mask >> 4 : mask) & 0xf;
 
          info->output_semantic[loc] = semantic + i;
+         info->output_semantic_to_slot[semantic + i] = loc;
 
          if (is_output_load) {
             /* Output loads have only a few things that we need to track. */
@@ -556,6 +558,8 @@ void si_nir_scan_shader(const struct nir_shader *nir, struct si_shader_info *inf
       info->tessfactors_are_def_in_all_invocs = ac_are_tessfactors_def_in_all_invocs(nir);
    }
 
+   memset(info->output_semantic_to_slot, -1, sizeof(info->output_semantic_to_slot));
+
    func = (struct nir_function *)exec_list_get_head_const(&nir->functions);
    nir_foreach_block (block, func->impl) {
       nir_foreach_instr (instr, block)

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 3e72e1fc37cc1b6a582b9be4b69e4b09d89d1f67..fb6ab7e0ee22b51c84f733ce8d3a43c4f481c2d7 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -3171,7 +3171,7 @@ static unsigned si_get_ps_input_cntl(struct si_context *sctx, struct si_shader *
                                      unsigned semantic, enum glsl_interp_mode interpolate)
 {
    struct si_shader_info *vsinfo = &vs->selector->info;
-   unsigned j, offset, ps_input_cntl = 0;
+   unsigned offset, ps_input_cntl = 0;
 
    if (interpolate == INTERP_MODE_FLAT ||
        (interpolate == INTERP_MODE_COLOR && sctx->flatshade) ||
@@ -3184,43 +3184,42 @@ static unsigned si_get_ps_input_cntl(struct si_context *sctx, struct si_shader *
       ps_input_cntl |= S_028644_PT_SPRITE_TEX(1);
    }
 
-   /* TODO: This search can be removed if we add a lookup table from semantic to index. */
-   for (j = 0; j < vsinfo->num_outputs; j++) {
-      if (semantic == vsinfo->output_semantic[j]) {
-         offset = vs->info.vs_output_param_offset[j];
-
-         if (offset <= AC_EXP_PARAM_OFFSET_31) {
-            /* The input is loaded from parameter memory. */
-            ps_input_cntl |= S_028644_OFFSET(offset);
-         } else if (!G_028644_PT_SPRITE_TEX(ps_input_cntl)) {
-            if (offset == AC_EXP_PARAM_UNDEFINED) {
-               /* This can happen with depth-only rendering. */
-               offset = 0;
-            } else {
-               /* The input is a DEFAULT_VAL constant. */
-               assert(offset >= AC_EXP_PARAM_DEFAULT_VAL_0000 &&
-                      offset <= AC_EXP_PARAM_DEFAULT_VAL_1111);
-               offset -= AC_EXP_PARAM_DEFAULT_VAL_0000;
-            }
+   int vs_slot = vsinfo->output_semantic_to_slot[semantic];
+   if (vs_slot >= 0) {
+      offset = vs->info.vs_output_param_offset[vs_slot];
 
-            ps_input_cntl = S_028644_OFFSET(0x20) | S_028644_DEFAULT_VAL(offset);
+      if (offset <= AC_EXP_PARAM_OFFSET_31) {
+         /* The input is loaded from parameter memory. */
+         ps_input_cntl |= S_028644_OFFSET(offset);
+      } else if (!G_028644_PT_SPRITE_TEX(ps_input_cntl)) {
+         if (offset == AC_EXP_PARAM_UNDEFINED) {
+            /* This can happen with depth-only rendering. */
+            offset = 0;
+         } else {
+            /* The input is a DEFAULT_VAL constant. */
+            assert(offset >= AC_EXP_PARAM_DEFAULT_VAL_0000 &&
+                   offset <= AC_EXP_PARAM_DEFAULT_VAL_1111);
+            offset -= AC_EXP_PARAM_DEFAULT_VAL_0000;
          }
-         break;
+
+         ps_input_cntl = S_028644_OFFSET(0x20) | S_028644_DEFAULT_VAL(offset);
+      }
+   } else {
+      /* VS output not found. */
+      if (semantic == VARYING_SLOT_PRIMITIVE_ID) {
+         /* PrimID is written after the last output when HW VS is used. */
+         ps_input_cntl |= S_028644_OFFSET(vs->info.vs_output_param_offset[vsinfo->num_outputs]);
+      } else if (!G_028644_PT_SPRITE_TEX(ps_input_cntl)) {
+         /* No corresponding output found, load defaults into input.
+          * Don't set any other bits.
+          * (FLAT_SHADE=1 completely changes behavior) */
+         ps_input_cntl = S_028644_OFFSET(0x20);
+         /* D3D 9 behaviour. GL is undefined */
+         if (semantic == VARYING_SLOT_COL0)
+            ps_input_cntl |= S_028644_DEFAULT_VAL(3);
       }
    }
 
-   if (j == vsinfo->num_outputs && semantic == VARYING_SLOT_PRIMITIVE_ID)
-      /* PrimID is written after the last output when HW VS is used. */
-      ps_input_cntl |= S_028644_OFFSET(vs->info.vs_output_param_offset[vsinfo->num_outputs]);
-   else if (j == vsinfo->num_outputs && !G_028644_PT_SPRITE_TEX(ps_input_cntl)) {
-      /* No corresponding output found, load defaults into input.
-       * Don't set any other bits.
-       * (FLAT_SHADE=1 completely changes behavior) */
-      ps_input_cntl = S_028644_OFFSET(0x20);
-      /* D3D 9 behaviour. GL is undefined */
-      if (semantic == VARYING_SLOT_COL0)
-         ps_input_cntl |= S_028644_DEFAULT_VAL(3);
-   }
    return ps_input_cntl;
 }